skymanbp/cc-tree

Idea The most convincing pitch for cc-tree is its output. Collect a handful of real runs (one per preset) into docs/gallery.md — each with the input, a trimmed tree.md, the final report, and the rendered diagram. Suggested seeds: brainstorm — a research/eng topic with a non-obvious winner attack — critique of a short public document or RFC design — an option × trade-off × reversibility decision code-audit — a deliberately buggy sample file This is both documentation and the basis for "showcase" social posts. What's involved Run the presets, trim the outputs for readability, and assemble docs/gallery.md with links from the README. No code changes. Good first issue? Yes — especially for someone learning the plugin who wants to contribute their first run.

Idea Every run already streams a structured tree.json. A small, dependency-light exporter would turn it into shareable visuals — which also doubles as marketing (people post the pictures). Targets: Graphviz DOT — .dot for dot -Tsvg Mermaid — paste-into-Markdown flowchart Interactive HTML — a single self-contained file (collapsible radial/indented tree, verdict colours) The existing tools/gen_radial_tree.py is a good reference for the radial layout + verdict glyphs; this issue is about driving a renderer from real tree.json instead of a hard-coded demo tree. What's involved A tools/export_tree.py reading tree.json and writing one of the formats. Start with DOT or Mermaid (smallest); HTML viewer is a nice stretch goal.

Idea --field <name> applies domain-aware reviewer weighting (feeds framings §3.C/§3.D/§3.I/§3.J). Today the set of shipped profiles is small. Adding more makes attack / code-audit / design noticeably sharper out of the box. Proposed new profiles (one field-profiles/<name>.md each, from field-profiles/_template.md): security — threat-model lens (STRIDE, trust boundaries, authz) ml — data leakage, eval validity, distribution shift, reproducibility frontend — a11y, perf budgets, state-management foot-guns distributed-systems — partial failure, idempotency, consistency, backpressure What's involved Each profile is one markdown file following the template; non-blocking if absent, so they're independent and low-risk. Great for anyone with domain expertise who wants to contribute without touching the engine. Good first issue? Yes — pick one domain you know well and write a single profile.

Idea Add a fifth shipped preset for literature / prior-art review — point cc-tree at a topic or a paper and have it grow a tree of what already exists, scored for relevance and novelty-gap. This is a natural sibling to brainstorm (which generates new directions) — research maps the existing landscape so brainstorm/design can build on it (great chaining target: research → brainstorm → design). What's involved A preset is a single presets/<name>.md file (see docs/presets.md). Concretely: pick the verdict vocabulary (e.g. KEY-PRIOR-ART / RELATED / TANGENTIAL / NEEDS-MORE-INFO) map the 12 framings to literature-review flavours (§3.X external-check should hunt prior art / surveys / citations) define the final report artifact (e.g. landscape.md) the schema is CI-enforced by tools/validate_plugin.py, and tools/test_validate.py should still pass Good first issue? Yes — it's authoring one file against a documented, validated schema; no engine changes required.